clone-page

Warn

Audited by Gen Agent Trust Hub on Apr 14, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/lib/browser.mjs uses execSync to perform runtime installation of the playwright package and Chromium browser. This enables the execution of arbitrary shell commands on the host system.\n- [COMMAND_EXECUTION]: Chromium is launched with the --no-sandbox and --disable-web-security flags in scripts/lib/browser.mjs. Disabling these security boundaries increases the risk of host compromise when the agent interacts with potentially malicious external websites.\n- [COMMAND_EXECUTION]: The scripts/lib/init.mjs file attempts to read a script using a relative path that reaches three levels above its own directory (../../../extract-page-data/...). This behavior attempts to access files outside the skill's defined scope.\n- [COMMAND_EXECUTION]: The scripts/query.mjs utility is vulnerable to path traversal. The file command uses path.join on user-supplied relative paths without sufficient sanitization, which could allow the reading of arbitrary files on the system.\n- [EXTERNAL_DOWNLOADS]: The skill performs automated downloads of the Playwright library and Chromium binaries from official registries. Additionally, scripts/lib/assets.mjs downloads arbitrary assets (images, fonts, and SVGs) from target URLs identified on the cloned webpages.\n- [REMOTE_CODE_EXECUTION]: Dynamic execution of npm install and npx playwright install at runtime fetches and executes external code. This setup process lacks version pinning and occurs outside of standard package management workflows.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Ingestion: Untrusted data enters the agent context via browser injection scripts in scripts/inject/extract-skeleton.js and scripts/inject/extract-section-styles.js. Boundary markers: Absent. Capability inventory: Includes subprocess calls (execSync) in scripts/lib/browser.mjs, file writes in scripts/lib/assets.mjs, and network operations in scripts/lib/assets.mjs. Sanitization: Absent for extracted text content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 14, 2026, 12:59 AM